1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
|
// ----------------------------------------------------------------------------
// - Open3D: www.open3d.org -
// ----------------------------------------------------------------------------
// Copyright (c) 2018-2024 www.open3d.org
// SPDX-License-Identifier: MIT
// ----------------------------------------------------------------------------
#include "open3d/core/MemoryManager.h"
#include <benchmark/benchmark.h>
#include "open3d/core/CUDAUtils.h"
namespace open3d {
namespace core {
enum class MemoryManagerBackend { Direct, Cached };
std::shared_ptr<MemoryManagerDevice> MakeMemoryManager(
const Device& device, const MemoryManagerBackend& backend) {
std::shared_ptr<MemoryManagerDevice> device_mm;
switch (device.GetType()) {
case Device::DeviceType::CPU:
device_mm = std::make_shared<MemoryManagerCPU>();
break;
#ifdef BUILD_CUDA_MODULE
case Device::DeviceType::CUDA:
device_mm = std::make_shared<MemoryManagerCUDA>();
break;
#endif
default:
utility::LogError("Unimplemented device {}.", device.ToString());
break;
}
switch (backend) {
case MemoryManagerBackend::Direct:
return device_mm;
case MemoryManagerBackend::Cached:
return std::make_shared<MemoryManagerCached>(device_mm);
default:
utility::LogError("Unimplemented backend.");
break;
}
}
void Malloc(benchmark::State& state,
int size,
const Device& device,
const MemoryManagerBackend& backend) {
MemoryManagerCached::ReleaseCache(device);
auto device_mm = MakeMemoryManager(device, backend);
// Warmup.
{
void* ptr = device_mm->Malloc(size, device);
device_mm->Free(ptr, device);
cuda::Synchronize(device);
}
for (auto _ : state) {
void* ptr = device_mm->Malloc(size, device);
cuda::Synchronize(device);
state.PauseTiming();
device_mm->Free(ptr, device);
cuda::Synchronize(device);
state.ResumeTiming();
}
MemoryManagerCached::ReleaseCache(device);
}
void Free(benchmark::State& state,
int size,
const Device& device,
const MemoryManagerBackend& backend) {
MemoryManagerCached::ReleaseCache(device);
auto device_mm = MakeMemoryManager(device, backend);
// Warmup.
{
void* ptr = device_mm->Malloc(size, device);
device_mm->Free(ptr, device);
cuda::Synchronize(device);
}
for (auto _ : state) {
state.PauseTiming();
void* ptr = device_mm->Malloc(size, device);
cuda::Synchronize(device);
state.ResumeTiming();
device_mm->Free(ptr, device);
cuda::Synchronize(device);
}
MemoryManagerCached::ReleaseCache(device);
}
#define ENUM_BM_SIZE(FN, DEVICE, DEVICE_NAME, BACKEND) \
BENCHMARK_CAPTURE(FN, BACKEND##_100_##DEVICE_NAME, 100, DEVICE, BACKEND) \
->Unit(benchmark::kMicrosecond); \
BENCHMARK_CAPTURE(FN, BACKEND##_1000_##DEVICE_NAME, 1000, DEVICE, BACKEND) \
->Unit(benchmark::kMicrosecond); \
BENCHMARK_CAPTURE(FN, BACKEND##_10000_##DEVICE_NAME, 10000, DEVICE, \
BACKEND) \
->Unit(benchmark::kMicrosecond); \
BENCHMARK_CAPTURE(FN, BACKEND##_100000_##DEVICE_NAME, 100000, DEVICE, \
BACKEND) \
->Unit(benchmark::kMicrosecond); \
BENCHMARK_CAPTURE(FN, BACKEND##_1000000_##DEVICE_NAME, 1000000, DEVICE, \
BACKEND) \
->Unit(benchmark::kMicrosecond); \
BENCHMARK_CAPTURE(FN, BACKEND##_10000000_##DEVICE_NAME, 10000000, DEVICE, \
BACKEND) \
->Unit(benchmark::kMicrosecond); \
BENCHMARK_CAPTURE(FN, BACKEND##_100000000_##DEVICE_NAME, 100000000, \
DEVICE, BACKEND) \
->Unit(benchmark::kMicrosecond); \
BENCHMARK_CAPTURE(FN, BACKEND##_1000000000_##DEVICE_NAME, 1000000000, \
DEVICE, BACKEND) \
->Unit(benchmark::kMicrosecond);
#ifdef BUILD_CUDA_MODULE
#define ENUM_BM_BACKEND(FN) \
ENUM_BM_SIZE(FN, Device("CPU:0"), CPU, MemoryManagerBackend::Direct) \
ENUM_BM_SIZE(FN, Device("CPU:0"), CPU, MemoryManagerBackend::Cached) \
ENUM_BM_SIZE(FN, Device("CUDA:0"), CUDA, MemoryManagerBackend::Direct) \
ENUM_BM_SIZE(FN, Device("CUDA:0"), CUDA, MemoryManagerBackend::Cached)
#else
#define ENUM_BM_BACKEND(FN) \
ENUM_BM_SIZE(FN, Device("CPU:0"), CPU, MemoryManagerBackend::Direct) \
ENUM_BM_SIZE(FN, Device("CPU:0"), CPU, MemoryManagerBackend::Cached)
#endif
ENUM_BM_BACKEND(Malloc)
ENUM_BM_BACKEND(Free)
} // namespace core
} // namespace open3d
|
